the catalogue's fastest win: what an empty Main Deck actually costs

Issue #200, 2026-09-13. Constructed, Duel (485) unless a Match (486) is named. Rules version 2026-07-16. Card text verbatim from data/corpus_flat.txt; rules pasted from data/Riftbound-Core-Rules-2026-07-16.txt. Read the unopposed clock first — every turn number below is measured against it.

Subject: renata-mastermind-points, which npm run adversarial -- --turns reports at T4 against a T6 baseline — the fastest point-scoring finisher in the catalogue and the largest margin in the table. It is an INFINITE that scores directly, standing on SFD-088 Renata Glasc, Mastermind, the only card in the pool that converts resources into a point inside the Main Phase.

This play walks it turn by turn and finds that T4 is not a floor, it is an artifact. Two costs the clock cannot see push it to T7, and the second of them is a precondition that the whole INFINITE class shares and that no instrument in this project prices.

1. The clock funds fuel downward and never upward

scripts/adversarial-check.mjs folds the needs/produces DAG in one direction only. An entry that produces only fuel gets its cheapest legal consumer added to its cost (the [fuel only: + …] column). An entry that consumes fuel is priced over its own uses[] alone, and its declared needs[] are never funded.

Measured AS OF the morning of 2026-09-13, over the 54 rows the clock covered then — the clock covers 80 now, because the ALT_WIN class was excluded from it by a bare class filter until later the same day, and on today's population the same two counts read 11 and 21. The counts move; the argument does not, so they are dated rather than chased. 8 rows produced points and declared unfunded needs — renata-mastermind-points, renata-bubble-bot-ready, jayce-mesmerize-renata, dragonstorm-brambleback-trinity-conquer, time-warp-hold-burst, swain-double-conquer, bottled-constellation-time-warp, reveler-loop-nasus-brambleback-conquer. Of the 9 rows that beat their baseline, three are in that set.

src/combos.ts generateVariants already computes the honest card set — it expands needs transitively and caps the union of domains at two (103.1.b). For this entry it returns exactly one variant:

renata-mastermind-points + lux-infinite-energy + lux-infinite-power — mind/order, 9 cards, 11 copies: SFD-088×1 OGN-104×1 SFD-180×1 SFD-166×1 OGN-212×1 OGN-110×1 UNL-165×2 UNL-173×2 OGN-087×1. Total E27 + 1 Power.

Running the clock's own exact allocator over that set gives deploy T5, last unit T4, so the honest figure is T5, not T4. Same correction on its two siblings: renata-bubble-bot-ready T4 → T5, jayce-mesmerize-renata T5 → T6, and lux-infinite-power T3 → T4.

That is a one-turn correction, and it is the small half of the problem.

2. The cost no instrument in this project prices: the deck has to be empty

lux-infinite-energy states it in its own first prerequisite:

"Main Deck EMPTY — the deck draws itself out first; every card not in the loop must be in hand or trash."

It is not one entry's quirk. Searched over prerequisites.notable for the exact string "the Main Deck is empty by construction": 14 of 14 INFINITEs carry it, and no entry of any other class does.

And ASSERTING it is not the same as NEEDING it — a first draft of this play conflated the two and the corrected version is the better finding. That sentence reaches all fourteen because it lives inside the Tournament Rules 505 paragraph that was pasted uniformly onto the class in #191 batch 27. Classified instead from each entry's own steps — does the pass put a card into the Main Deck and draw it back? — the class splits:

• Eight need it. lux-infinite-energy, lux-infinite-power, lady-luminosity-loop-comet, jhin-virtuoso-ekko-malzahar-vi and gemdragon-henge-vi-blind-fury recycle and draw inside every pass; renata-mastermind-points, renata-bubble-bot-ready and jayce-mesmerize-renata declare needs: ["infinite-energy","infinite-power"] and inherit it from the Lux engines.
• Six do not. jhin-fiora-facebreaker-recall, twilight-reveler-eye-facebreaker-recruits, garen-fiora-malzahar-facebreaker-recruits, threshold-reveler-infinite-energy, reveler-svellsongur-jhin-infinite-power and the banned pursuer-herald-recruits are combat-recall loops that spend zero cards a pass — walk in, take the Attacker designation, get recalled by 466.1.a.2, repeat. Two of them do carry the word draw in their steps and both say "first pass only"; their own netPerIteration reads "0 Energy, 0 Power and 0 cards spent". For those six the boilerplate's argument is not merely unnecessary, it is the wrong one: 416.5 never fires because nothing is recycled, and saying the deck is empty by construction claims a precondition they do not have.

So of the 9 rows that beat their baseline that morning, five carried the unpriced empty-deck cost (lux-infinite-power, renata-mastermind-points, renata-bubble-bot-ready, lux-infinite-energy, jayce-mesmerize-renata), one was a CHAIN (yasuo-windrider-ride-the-wind-chain), and three were honest: jhin-fiora-facebreaker-recall and twilight-reveler-eye-facebreaker-recruits at T5, with garen-fiora-malzahar-facebreaker-recruits tying at T6.

Re-measured that evening, once ALT_WIN was in the clock and needs was folded upward: 21 rows beat their baseline and FOUR of them carry the empty-deck cost — the same four minus jayce-mesmerize-renata, which the fold moved to T6 and out of the set. Twelve of the twenty-one are ALT_WINs, which nobody could see this morning. Re-derive these, never quote them; what does not move is the 8-and-6 split above, because it is a property of the entries rather than of the table.

Those three are also the inverse of the line this play walks. They are combat loops: 807.1.d needs the Attacker designation and 323.9 stages no Combat without an enemy garrison, so they are dead on an empty board and alive on a contested one — which is the shape the --stalled pass calls a credential, and the shape the Renata line does not have. The catalogue's genuinely fast INFINITEs are the ones that cost no cards and want the board a stall produces.

The empty deck is a ONE-MAIN-PHASE state, and it destroys itself

Three paragraphs, none of which any entry cites:

315.4.b.1. "If there are no cards remaining in their Main Deck to draw, the Turn Player has been Burned Out."

431.2.b. "Recycles their trash into their Main Deck." · 431.2.c. "Chooses an opponent to gain 1 point."

315.4.b.2. "After completing the Burn Out the Turn Player still Draws 1."

So an empty Main Deck cannot be carried across a turn boundary. Your next Draw Phase Burns you Out automatically: the opponent gains a point, your whole trash goes back into your deck — randomized by 431.2.b's own reminder — and the loop's premise is gone. The Draw Phase (315.4) sits before the Main Phase (316), so you must empty the deck and win in the same Main Phase.

The consequence for the ledger is exact rather than approximate: the kill turn must open its Beginning Phase with more than zero cards and open its Main Phase with zero. With the mill below that means exactly 4 cards at the start of the kill turn — Burn 3 in the Beginning Phase, the Draw Phase takes the last one.

Mind/Order has no mill of its own

Swept over data/corpus_flat.txt for [Burn N]: eight printings, and seven are Fury or Chaos (VEN-002, VEN-022, VEN-095, VEN-108, VEN-113, VEN-114, VEN-144). The eighth is colourless and is the only one a Mind/Order deck can run:

VEN-165 Shadow Temple — Battlefield, Colorless — "When you hold here, [Burn 3]. (Put the top 3 cards of your Main Deck into your trash.)"

And the pool's one Skip-your-Draw-Phase card, VEN-022 Endless Riches, is Fury — so there is no legal way for this deck to dodge 315.4.b.1 either.

Which makes the plan's engine and its fuse the same card. 315.2.b.2 "The Turn Player Holds all Battlefields they Control" is mandatory and Shadow Temple's clause carries no "may", so once you control it you burn three a turn whether you want to or not — and by 431.1.b a mandatory Burn 3 with fewer than three cards left is itself a Burn Out. The landing has one legal value.

3. The deck, and why 21 of its cards exist only to be thrown away

Chosen Champion: SFD-088 Renata Glasc, Mastermind. She is a unit and is not a Signature card (signature: false in cards.json; 103.2.d.3 bars Signature cards from the Champion Zone), so 103.2.a.1 puts her in the Champion Zone at the start of the game and 108.3.d lets her be played from there. She never has to be drawn, and the Main Deck is 39.

Battlefield: VEN-165 Shadow Temple. In a Duel, 485.5 "randomly selects" one of your three, so it is one in three. In a Match, 486.5 says only "Each player selects one (1) of their three" — no randomness in game 1 — and then removes it for the rest of the match if you won with it.

Main Deck 39, of which the plan below uses 18:

Domain identity: every card is mono-Mind or mono-Order, Shadow Temple is colourless, so the union is {mind, order} and 103.1.b admits any of the four Mind/Order legend names — OGN-265 Herald of the Arcane, OGS-021 Lady of Luminosity - Starter, SFD-201 Chem-Baroness, UNL-199 Deceiver.

4. The mana, so the table is checkable

A rune pays 1 Energy (164.2.a, cost = its exhaust) and recycles for 1 Power of its domain (164.2.b, cost = the recycle, no exhaust), but a recycled rune leaves the board for the Rune Deck (161.2.b) and comes back at 2 a turn (315.3.b), capped at 12 (161.2.a). Going first — 485.7 gives the extra rune to the player going second, so this is the slower seat.

The loop needs 11–12 runes on the board and that is a hard floor. Per pass it spends 10 Energy and 1 Mind Power (Ekko 5, Shadow's Call 2, Sacrifice 1, Forge 2) and Ekko's Deathknell readies the runes once, so the steady income is R − 1 Energy per pass against a spend of 10: R = 11 breaks even, R = 12 nets +1. Since R = 2T minus every Power ever spent, 12 runes is turn 6 at the earliest and every pre-emptive Power spend pushes it later — which is why the walk below buys no Power-costed draw at all and skips OGN-114 Progress Day (E6 + 1 Power) and SFD-087 Premonition (E2 + 3 Power) entirely.

5. The turns

D = cards left in the Main Deck. 39 − 4 opening (rule 116, "Players each draw 4") = 35.

T4, the seeding turn, in full

The turn that pays for itself, because Ekko's Deathknell readies the runes mid-turn:

1. Tap all 8 runes → 8 Energy. Recycle one already-exhausted Mind rune → 1 Mind Power (164.2.b costs the recycle, not an exhaust). Board: 7.
2. OGN-110 Ekko, Recurrent — E5 + 1 Mind Power. 3 Energy floating.
3. UNL-165 Shadow's Call E2 on Ekko: "Choose a friendly unit without [Temporary]. Give it [Temporary]. Draw 2." 1 Energy floating.
4. UNL-173 Sacrifice E1 — its additional cost kills a friendly [Mighty] unit, and Ekko is printed Might 5.
5. Ekko's Deathknell resolves first (340.1, newest finalized item): "Recycle me to ready your runes." Ekko goes to the bottom of the Main Deck and all 7 runes are ready again.
6. Sacrifice resolves: draw 2, channel 1 rune exhausted → board 8.
7. Tap the 7 ready runes → 7 more Energy: OGN-212 Forge of the Future E2 (its play trigger makes a 1 Might Recruit at your base) and OGN-083 Consult the Past E4 (draw 2).

Net runes: zero. One recycled for Ekko's Power, one channelled back by Sacrifice. That is the whole reason the 12-rune mark still lands on turn 6. And the turn leaves exactly what the loop needs in the trash: one Shadow's Call and one Sacrifice.

Card accounting, which closes exactly

25 cards drawn (4 opening + 7 Draw Phases + 2 + 2 + 6 + 4), 15 burned by five Holds, one Ekko recycled back in: 25 + 15 = 40 = 39 + 1. The deck reaches zero with nothing left over, which is the only way to arrive at an empty Main Deck without 431.1.a doing it for you.

T7, the kill

R = 12, D = 0, 6 points on the board, 2 needed. Hand: Ekko, Shadow's Call #2, Sacrifice #2, Retreat, Rally the Troops, Lecturing Yordle. Board: the Sentry at Shadow Temple, Fiora, the Forge.

1. Kill the Forge (free): "Recycle up to 4 cards from trashes" — take Forge, Shadow's Call #1, Sacrifice #1 (416.6 gives the choice to the recycling player). D: 0 → 3.
2. Run the pass of §4. D: 3 → 1 (Shadow's Call draws 2) → 2 (Ekko's Deathknell recycles him) → 0 (Sacrifice draws 2). Hand is identical, board is identical, +1 Energy. Unbounded.
3. Convert Energy to Power with lux-infinite-power — Forge, Retreat and Lecturing Yordle, at 9 Energy per Power.
4. SFD-166 Rally the Troops E2: "When a friendly unit is played this turn, buff it."
5. Play Renata from the Champion Zone (E5) to Shadow Temple — 355.2.a: "Valid locations include the controller's Base or a Battlefield the controller controls." Rally buffs her from Might 4 to 5, which is 709's crossing, so SFD-180 Fiora, Worthy pays 1 Order Power to ready her. She entered exhausted (143.4) and her Score ability costs an exhaust, so without Fiora she could not fire at all (414.1.b with 203.3).
6. E4 + 4 Mind Power, exhaust: score 1 point (7).
7. OGN-104 Retreat E1 returns her to hand; replay her E5. 705 stripped the buff when she left play and 124 makes her a new object, so Rally buffs her again, she crosses 5 again, Fiora readies her again. Score (8). Win on turn 7.

6. Breaks to

• OGN-133 Flurry of Blades — Body, E1, [Reaction] — "Deal 1 to all units at battlefields." The Sentry is Might 1. Killing it strips Control of Shadow Temple at the next Cleanup (323.6), which takes the mill, the Hold points and the place Renata has to stand, all for one Energy. The entry declares zone: BASE on nothing here, so this is the catalogue's known hole arriving in the one place where it costs three things at once. Garrison with a Might-2 body instead and it costs one more Energy on turn 1.
• Anything that contests Shadow Temple. Not removal — just a body. 344.2 needs "there aren't units controlled by different players there", so an opposing garrison turns the Conquer into a Combat, and a lost Combat costs the same three things.
• The burns are not yours to choose. Fifteen cards go to the trash off the top of the deck, and any of the eighteen plan cards can be among them. The Forge's first activation retrieves four of them, and beyond that the answer is 103.2.b's three copies of the cheap pieces. This walk is a best case and assumes no plan card is buried.
• One turn of disruption is fatal in a way it is not for other lines. Miss the T7 landing and the next Draw Phase Burns you Out: opponent +1 point, trash shuffled back into the deck, and the setup starts again from a deck that is no longer 39 cards of your choosing.

7. The verdict, and it has three cases rather than one

The line's mill, its points and its payoff's legality are all the same requirement — control of one battlefield — and so is the free Hold curve's. A finisher earns its slot where the Hold curve has stalled, and this one stalls on exactly the same condition. That is why the honest answer to "is it viable" is yes, in the middle case, and why T4 was never the right number to compare against T6.

8. What writing this found in the catalogue's own instruments

The BOARD-INDEPENDENT bucket is half wrong, and the label overclaims

npm run adversarial -- --stalled classifies every finisher by what the board must look like. Its last bucket, INDEPENDENT, holds 8 rows under this label:

"BOARD-INDEPENDENT. No Hold, no Conquer and no attack in the printed text, so nothing about the board switches it off."

It is the else branch: three regexes over card text and the entry's own steps for hold, conquer and attack, and anything matching none of them lands here. There is no predicate for a location gate, and prerequisites is never read at all. So:

• Four of the eight stand on SFD-088 Renata Glasc, Mastermind, whose own printed text ends "Use my abilities only while I'm at a battlefield" — renata-mastermind-points, renata-bubble-bot-ready, jayce-mesmerize-renata, jhin-virtuoso-ekko-malzahar-vi. With 355.2.a that is a requirement to control a battlefield, which is precisely what a stall removes.
• Two more are UNL-088 Gutter Palace: "if you have exactly 4 cards in hand and exactly 4 units at battlefields, you win the game." Not a control requirement, but not board-independent either.
• One (lux-infinite-power) scores no points at all; it is fuel.

That leaves exactly one finisher in the catalogue that both scores and needs nothing whatever from the board: bottled-constellation-time-warp — VEN-067 Bottled Constellation, "At the start of your Main Phase, you may kill 3 other friendly units and/or gear to score 1 point", which works from your base. The same script prices it at T13 against a T6 baseline, the slowest row in the table.

This is the precedence lesson this project already recorded for the Conquer-before-Hold ordering, arriving in a new place: it is the residual bucket that inherits everything the patterns cannot see, and its label is the one that should be written most cautiously. renata-mastermind-points says "You control a battlefield to play Renata to" in its own prerequisites.easy; the classifier reads uses[] and steps and never looks.

Two of the seven entries using SFD-088 say nothing about the battlefield anywhere — renata-bubble-bot-ready and jayce-mesmerize-renata — and both are in that bucket.

The project ruled on this clause two months ago and never applied it to Renata's own lines

R30 (voted 2026-09-06, issue #11) turns on exactly this sentence: Heimerdinger copying Renata's exhaust: Score 1 point copies "Use my abilities only while I'm at a battlefield" with it, so Heimerdinger must stand at a battlefield. The reading was settled for the copier and never read back onto the original — the diagnosed-but-unapplied shape, in a new place.

The Tournament Rules 505 notable is right on eight entries and wrong on six

Every INFINITE carries the same pasted paragraph, and its load-bearing clause is "The condition is checkable and this entry meets it: the Main Deck is empty by construction, so every card recycled in a pass is drawn back inside that same pass." On the eight loops that recycle, that is exactly right and is one of the better things in the catalogue — it is the "zero spare draws" invariant doing a second job, licensing the shortcut under TR 505.9.

On the six combat-recall loops it is an argument for a condition they do not have. The true reason those six may be shortcut is shorter and stronger: they recycle nothing, so 416.5 never fires at all. Uniform boilerplate is how a correct sentence ends up on an entry it was not measured against, and the tell here is that the entries' own netPerIteration fields contradict it in their own words — "0 Energy, 0 Power and 0 cards spent".